Skip to content

PromptPotter/prompt-potter-optimizer / .claude/skills/potter-run

potter-run

Verified

Launches (`new` / `resume`), supervises and diagnoses PromptPotter campaigns, and onboards a new dataset. Fires on "/potter-run", "start a campaign", "run promptpotter-self", "resume", "watch it", "how's the run", "why did round N do that", a dataset named to optimize, or a campaign already in flight that needs supervising.

FileKindSize
SKILL.mdskill md6.0 KB
reference/onboarding.mdmarkdown2.8 KB

reference/onboarding.md

View on GitHub
# Onboarding a dataset that has never run

No campaign, no loader, or a cold machine. A registered benchmark skips all of it: `new <name>`.

## The flow — upload → context → origin → select · modify · start

One check-in, one origin, no hand-written loader. Web: `#/new` → upload → a durable draft at `#/d/<draft id>`
→ one `checkin` turn proposes column map, prompt fields and `task_context` → a checklist gate (no LLM)
blocks mint until query + ground_truth + framing are CONFIRMED and every active LLM node owns a model → C0
→ the picker → mint. The CLI rides the same seam:

```bash
python -m promptpotter new <file.csv> --set raw_task_description='what the prompt does'
```

Omit `--set` to let the resolver propose the framing and ask; `--set` takes `EditDraftPatch`'s own
dotted paths: `--set config.optimization.optimizer=capo` and
`--set config.optimization.nodes.<node>.config.<knob>=V` are the picker's answers. A run stopping short
of a complete origin keeps its check-in; Discard is `discard-draft <draft_id>`.

## Fast path — Claude-simulated check-in

When the operator says the data is ready **and it loads cleanly**, author the origin yourself:

1. **Test the load first**: raw file → `POST /commands/ingest-upload` (a form); a registered dataset →
   `POST /commands/open-dataset-draft`; a prior origin → `POST /commands/open-origin-draft`.
   **If the load fails, do NOT simulate.**
2. **Author the origin** via `POST /commands/edit-draft-campaign`: prompt fields, `task_context`, the
   column map, a plain `task_description`. The closed-label answer space stays code-owned.
3. **Gate, then start**: `origin_readiness` complete, then `POST /commands/start-checkin` with `{draft_id}`.

## Cold machine

- **Missing `.env`.** Ask for `GROQ_API_KEY`; add OpenAI/Anthropic/OpenRouter only if named. `.env.example` is the template.
- **Backend unreachable.** TermNorm is the canonical test backend: `git clone
  https://github.com/runfish5/TermNorm-excel` beside this repo, record it as `termnorm_checkout` in
  `CLAUDE.local.md` (`promptpotter/connectors/CLAUDE.md` § TermNorm is not a third party), start
  `start-server-py-LLMs.bat` (the autonomy mode says who), wait for `/status` 200.
- **No loader.** A tenant dataset is ingested, never given a loader. A new fetched benchmark (rare): a
  field mapping in its `source:` on the `columns` recipe, never code, and a dataset folder drafted from
  `justlogic/d234`, per `docs/operations/datasets.md` § Adding a dataset.

## First-run smoke

If `{name}` has never produced a measurement, one round (it mints, so the autonomy mode decides):
`python -m promptpotter new {name} --max-rounds 1 --set config.optimization.nodes.l1_generate.config.n_variants=3 --set config.optimization.nodes.adaptive_queue.config.sp_budget_round=5 --set config.optimization.nodes.escalation.config.l1_patience=1`.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall